Use when planning, executing, diagnosing, verifying, hardening, or rolling back deployments and runtime operations, including releases, Docker Compose, containers, systemd, Kubernetes, health checks, incidents, environment configuration, production or staging troubleshooting, and local development runtime startup.
Use when a GitHub issue needs to be created for a product goal, exploration, prototype, bug, feature, architecture change, optimization, learning, documentation, security, or tech-debt item.
Use when working on one or more existing GitHub issues in a repository and the agent must recover goals, evidence, decisions, implementation state, or delivery work.
Use when a user describes a product requirement, feature idea, user need, or ambiguous outcome that should be clarified, evaluated, or persisted in GitHub.
Use when rewriting or reviewing Chinese or English prose that sounds AI-generated, formulaic, overly polished, promotional, generic, or unlike the author's voice, including everyday writing, technical documentation, and pull request descriptions.
Use when faithfully implementing a web interface from a supplied screenshot, Figma frame, approved visual reference, or requested exploratory reference.
Use when the user asks to install a proxy, set up Mihomo, configure Clash subscription proxy, or set up proxy wrappers for dev tools (Cursor, Claude, Copilot). Handles full lifecycle — install, configure, verify, and uninstall — in pure user-space (no root required).
Use when reading, creating, editing, or validating a Microsoft PowerPoint .pptx or .potx artifact, including repository-backed weekly, monthly, project-status, architecture, MVP, release, and executive report decks.
Use when the user explicitly requests production or live business and operational statistics from a database and the work must remain read-only. Do not use for development, debugging, staging data, deployment, health checks, schema migrations, data repair, or any data-changing task.
Use when an engineering task involves writing, changing, fixing, refactoring, debugging, optimizing, or shipping code and the work must stay aligned with user goals while choosing evidence proportionate to uncertainty and risk.